The accuracy of the model (MLP) decreased when I removed duplicated features (two columns) from dataset. Should I remove them or keep them? It is supposed to get improved, any idea why this happened?
If the features are 100% the same they can result in biased training e.g. imbalanced dataset. Yes, J. Rafiee
is right it's hard to say if the features are exactly the same. So, check on the results, if using all features (i.e. the similar one too) the NN results in biased training then try removing them.
Of course if you remove your duplicate features then your accuracy will decrease because if you have duplicate values then it will easy for your model to predict and accuracy will increase,
try to clean the data using powerful techniques or preprocessed to balanced training data. besides, you can apply pruning techniques to remove redundant features safely without big damage in accuracy.
Preparing Data is very important step in any MLP.In your project ACC decreased for removing the duplicated features ,this means that these features aren't same and important in model fitting .
When I checked the duplicated features, I found them 100% exactly the same. However I am not using MPL as a final model. It is used for evaluating feature selection techniques.
It was selected because my classification problem is nonlinear.
See if you can use Deep learning-based networks as they involve automatic feature extraction rather than hadncrafted features. So there won't be any confusion of duplicate features.